{"id":"W4407721618","doi":"10.1177/23821205251321791","title":"Program Evaluation in Competence by Design: A Mixed-Methods Study","year":2025,"lang":"en","type":"article","venue":"Journal of Medical Education and Curricular Development","topic":"Innovations in Medical Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Competence (human resources); Descriptive statistics; Medical education; Program evaluation; Specialty; Subspecialty; Program Design Language; Curriculum; Psychology; Computer science; Medicine; Family medicine; Political science; Pedagogy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009231816,0.0001134134,0.0002821921,0.000456483,0.00006840986,0.0000301826,0.0001488527,0.0001225847,0.0002033229],"category_scores_gemma":[0.0051026,0.00008749363,0.00002897361,0.0009554661,0.00006413808,0.00007907266,0.00003673376,0.000470049,0.00000214665],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000388448,"about_ca_system_score_gemma":0.006432636,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000006589567,"about_ca_topic_score_gemma":0.000002244213,"domain_scores_codex":[0.9969115,0.0004578794,0.0009179547,0.0001716959,0.001395337,0.0001456687],"domain_scores_gemma":[0.9986059,0.0000793907,0.000241843,0.0001190168,0.0007944827,0.0001593469],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00001984195,0.005520817,0.01637613,0.00004933244,0.00005278239,0.000003722898,0.001335216,0.000001627555,0.00002611904,0.00009827584,0.02705137,0.9494647],"study_design_scores_gemma":[0.00759902,0.0009804076,0.6448675,0.004041239,0.0003793377,0.0004190066,0.03388243,0.01075489,0.0007101663,0.000787739,0.2952271,0.0003511767],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9303187,0.003127219,0.04764564,0.01459696,0.002103661,0.001932612,6.288852e-8,0.00001546582,0.0002597517],"genre_scores_gemma":[0.6692106,0.000194502,0.3262146,0.003461679,0.000135715,0.000572193,0.0000199944,0.00001034451,0.0001803886],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9491136,"threshold_uncertainty_score":0.9992,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02647637743565949,"score_gpt":0.4709311998325691,"score_spread":0.4444548223969096,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}